Build an AI system that solves research problems, then charge people to use it. That is how the scientific AI business has generally been understood. But as general-purpose AI takes on a wider range of problems, specialized services face a difficult question: if customers can get similar answers elsewhere, what will they keep paying for?
This is not to say that general-purpose AI will replace every specialized calculation or prediction. Even so, the ability to solve a particular problem may not provide a lasting advantage on its own. What becomes important, then, is access to experiments that test those predictions—and who owns the results that follow.
Let us revisit the companies covered in this issue from that perspective.
There Is Still Work to Do After AI Gives an Answer
An AI system’s proposal of a promising molecule does not, by itself, make a drug. Can the molecule actually be made? Does it have the intended effect? How does it behave in the body? Beyond the prediction lies information that only experiments can provide.
Ginkgo’s announced partnership with TuneLab turns this work into a service. It provides experiments for companies using AI and returns the results as data suited to machine learning. Ongoing work arises not only for those selling predictions, but also for those testing them. Ginkgo’s announcement ↗
This suggests that advances in AI will not necessarily reduce the value of experiments. If generating hypotheses becomes easier, the next scarce capability may be determining which of them will actually work.
Owning experimental equipment does not automatically create an advantage, however. Can a company repeatedly deliver the measurements customers need, with dependable quality and costs? Do its contracts allow the resulting data to be used in future work? Whether access to experiments becomes a competitive advantage depends on the details.
If the Answers Are That Promising, Why Not Profit From Them?
Taking the argument a step further raises another question.
Suppose you genuinely believe your technology can identify valuable drug candidates. Why collect only a fee for using the tool? Why not hold rights to the candidates themselves and share in the returns if they succeed?
The transaction between Schrödinger and Tectora puts that idea into practice. Schrödinger contributed two drug discovery programs and received equity and rights to future milestone and royalty payments. Rather than only licensing its software for others to use, it also participates in the value of the resulting assets. Schrödinger’s official announcement ↗
Of course, identifying a promising candidate is different from successfully developing it. Holding a greater claim on the outcome also means taking on financial commitments and the risk of failure. This is a reason to work with investors and development companies, sharing both the returns and the burdens rather than shouldering everything alone.
That is what makes these agreements interesting. The way a company receives its returns reveals where it sees value in its technology—and how much risk it is willing to take itself.
What Matters Is What the Platform Actually Provides
It is too early to read these developments as scientific AI companies collectively abandoning platform businesses. Back in 2019, Schrödinger was already outlining plans to expand its own and collaborative drug discovery programs. Owning the results is not, in itself, a new strategy. Official announcement from 2019 ↗
Nor has it been established that concern about general-purpose AI prompted the companies’ decisions discussed here.
Nevertheless, it is worth reconsidering these developments against the backdrop of AI progress. A business that provides only a model or an interface faces different competitive conditions from one that can obtain proprietary experimental data, advance real-world development, or hold rights to successful outcomes.
A platform is a way to organize a business, not an explanation of its competitive strength. What does it provide that keeps customers from leaving? What does it have that customers cannot get elsewhere? And what share of the value created for customers or society can it earn as profit?
We believe competition in scientific AI will increasingly require measures beyond the difficulty of the problems a system can solve. As the ability to produce answers becomes more widely available, companies that can test those answers, put them into practice, and turn them into results may have a more important role to play.
We want to track more than the number of AI partnerships. Who holds the rights to experiments and data? Who pays for development? Who receives the returns when it succeeds? Watching how those arrangements change can reveal whether companies are trying to sell AI—or use AI to create another kind of value.